Tool methodology
Tool Methodology
This page documents the sources, thresholds, and limits behind the Reference Checker, Bibliography Alphabetizer, and Journal Abbreviation Converter.
Last reviewed: August 29, 2026
Claims and evidence stay separate
What you paste is a claim, not a fact. The Reference Checker extracts a claimed title, author, year, and identifier from your entry first, without touching the network. It resolves evidence from Crossref and PubMed second, and compares the two objects third. Evidence can confirm or contradict a claim. It never silently overwrites the claim before that comparison happens.
Evidence thresholds
The Reference Checker scores how closely a claimed title matches the closest database record. A similarity of 0.75 or higher returns a verified verdict. A similarity from 0.40 up to that point returns a mismatch, with the closest record attached as evidence. Anything below 0.40 returns not found, unless an exact identifier already produced a mismatch.
The bare identifier exception
A reference that is only a bare DOI or a bare PMID, with no title to compare, can still verify. In that case the checker resolves the identifier directly against the DOI Foundation's DOI Handbook or PubMed, and the verdict states that identifier resolution was its evidence basis rather than a title match.
Completeness rules
A verified or mismatch verdict from the Reference Checker always carries the claimed title alongside the matched title and, where available, a DOI link back to the source record. A generated Journal Abbreviation Converter result stays visibly labeled as generated rather than authoritative. The Bibliography Alphabetizer never deletes or rewrites an entry: it only reorders the entries you gave it and flags the ones it is uncertain about.
Authority fallback
The Reference Checker resolves a claimed DOI against Crossref first. If Crossref has no record for that DOI, it searches Crossref by title, and, for references with a biomedical hint, also searches PubMed by title, to check whether the reference exists under a different DOI. The Journal Abbreviation Converter checks NLM's Citing Medicine catalog first, then a versioned CC0 open journal-title dataset, then Crossref metadata, and only generates an ISO 4 abbreviation from LTWA rules when none of those sources has an exact record.
Batch limits
The Reference Checker accepts up to 30 references per request; anything past that is not sent. The Bibliography Alphabetizer accepts up to 100 entries, 1,500 characters per entry, and 150,000 characters total, all inside your browser. The Journal Abbreviation Converter accepts up to 300 characters per lookup.
Data handling and privacy
The Bibliography Alphabetizer never leaves your browser: CiteTools does not receive the pasted bibliography. The Reference Checker and the Journal Abbreviation Converter send your input to the CiteTools server so it can query Crossref, PubMed, and NLM on your behalf. Neither tool stores what you paste after the request completes. Read the privacy policy for the full boundaries.
OpenAI fallback
The Reference Checker, Bibliography Alphabetizer, and Journal Abbreviation Converter form a claim-and-evidence pipeline that does not call OpenAI. CiteTools' separate citation converter calls OpenAI (gpt-4o-mini) to suggest missing fields when a pasted reference is incomplete, and only when an OpenAI key is configured on the server.
Cache lifetime
Journal Abbreviation Converter results are cached in server memory for up to 24 hours per input and standard combination, then re-resolved from the underlying sources. The Reference Checker and Bibliography Alphabetizer do not cache results between requests.
Rate limits
CiteTools rate limits its server operations using Vercel's firewall. The published rule allows 30 requests per minute per client for each operation, and one operation reaching that ceiling does not block a different operation from the same client. Vercel counts requests per region, not as a single counter shared across every region a request might hit, so a client that is rate limited in one region is not guaranteed to be rate limited everywhere. The Bibliography Alphabetizer runs in your browser and has no server-side limit.
Export and artifact validation
Where CiteTools generates a downloadable file elsewhere on the site, it checks the file's structure, such as its ZIP signature or balanced braces, before sending it. A malformed file is treated as a failure rather than returned as a success with broken content.
Known source coverage gaps
Crossref and PubMed index journal articles, conference papers, and preprints most completely. Books, theses, legal sources, and most websites are weakly covered or absent from both, so a not found verdict from the Reference Checker is not proof that a reference is fake for those source types; verify it manually instead. NLM's catalog and the CC0 open-title dataset do not cover every journal. When both miss, the Journal Abbreviation Converter falls back to a generated ISO 4 form, which is useful for review but is not an authoritative record.
Current editions
The Bibliography Alphabetizer orders APA, MLA, and Chicago entries using each style's current guidance. Harvard entries follow Cite Them Right ordering rules. Cite Them Right's 13th edition has been current since July 31, 2025. CiteTools' own Harvard citation style still renders the legacy CSL template, labeled Cite Them Right 12th Edition (Legacy) everywhere that label appears.
Accessibility
This page and its sibling tool pages target WCAG 2.2 AA contrast, keyboard access, and focus visibility.
Failures
A request that fails schema validation, such as too many references or a missing field, returns HTTP 400 and is not retried automatically. A rate-limited request returns HTTP 429 and can be retried after the window resets. An upstream lookup that times out or errors returns HTTP 502 or 504 and can be retried. Inside a batch, each reference or entry is judged on its own: one failed or unparseable item never fails the rest of the batch.